Back to Blog
Mate Benyovszky

Connect Any Playbook as an MCP Server — Cursor, Claude, and Beyond

Step-by-step guide to connecting AgentPlaybooks to Cursor IDE, Claude Desktop, Claude Code, and other MCP-compatible tools. One URL, copy-paste config, instant AI tools.

Connect Any Playbook as an MCP Server

Every playbook on AgentPlaybooks is also a live MCP (Model Context Protocol) server. That means you can plug it straight into Cursor, Claude Desktop, Claude Code, or any MCP-compatible client — and your AI agent immediately gets access to the playbook's tools, memory, canvas, and personas.

Today we're making this even easier with a new Integrations tab in the playbook dashboard and updated documentation covering Cursor IDE setup front and center.

Update (October 2026): three ways in

Since this post was written, a playbook can be reached three ways. Pick by scope and by client:

| You want | Use | Authentication | |---|---|---| | Your whole account — every playbook, create and manage them from the agent | https://apbks.com/api/mcp/manage | OAuth sign-in (the client opens a consent screen), or a user API key | | One playbook — exactly its tools, memory, and skills | https://apbks.com/api/mcp/YOUR_GUID | A playbook API key — or none, for a public playbook's read access | | Scripts and code-executing agents — no MCP at all | https://apbks.com/api/mcp/YOUR_GUID/llms.txt | Same playbook API key |

Account endpoint. One connection covers every playbook you can access; the tools take a playbook_id argument. It is a real OAuth 2.1 protected resource, so in Claude's Add custom connector dialog choose Always required and sign in — no key to copy. Headless automation can send a user API key instead.

Playbook endpoint. Bound to one playbook by its URL, authenticated with a playbook API key. The key may go in Authorization (with or without the Bearer prefix) or in X-API-Key — useful when a client reserves Authorization for itself. Append ?toolset=runtime (or memory, admin) to advertise fewer tools; on an unpinned connection the agent can still reach everything through find_tools.

Scripts. Some agents write code instead of calling tools one by one, and some of those cannot call MCP tools from inside their scripts. For them, llms.txt is a single fetch that explains authentication, the one-POST calling convention (POST .../tools/TOOL_NAME with JSON arguments), and gives every tool a one-shot example — after which the agent chains as many calls as it needs in one script, keeping intermediate results out of its context.

The dashboard's Integrations tab now shows all three, each with a copy button.

What is MCP?

The Model Context Protocol is an open standard (originally developed by Anthropic) that lets AI assistants connect to external tools and data sources over a simple JSON-RPC interface. Think of it as a universal plugin system for AI.

AgentPlaybooks implements the MCP server specification for every playbook. Your memory becomes readable/writable state, your skills are accessible via built-in tools (list_skills, get_skill), and your personas provide system prompt context — all via a single HTTP endpoint.

The New Integrations Tab

We've restructured the playbook editor. The old "API Keys" tab is now called Integrations and contains everything you need to connect your playbook to external platforms:

  • Connect as MCP Server — Ready-to-copy JSON configs for Cursor, Claude Desktop, and Claude Code. The configs are pre-filled with your playbook's GUID and name.
  • Use with AI Platforms — Quick-action buttons (Open in Claude, Open in ChatGPT, export as ZIP) plus API endpoint reference.
  • Platform Cards — One-click links to step-by-step guides for Cursor, ChatGPT, Claude, Gemini, Claude Code, and generic API integration.
  • API Keys — Generate and manage authentication keys for write access.

Connecting to Cursor IDE

Cursor has native MCP support. Here's how to connect:

1. Open the Integrations Tab

Go to your playbook in the dashboard and click the Integrations tab (the puzzle icon).

2. Copy the Cursor Config

You'll see a ready-to-copy JSON block. It looks like this:

{
  "mcpServers": {
    "apb-my-assistant": {
      "url": "https://apbks.com/api/mcp/YOUR_GUID"
    }
  }
}

3. Paste into Cursor Settings

Save the JSON to one of:

  • Project-level: .cursor/mcp.json in your project root
  • Global: ~/.cursor/mcp.json for all projects

4. Restart and Verify

Restart Cursor (or reload the window). Your playbook's tools will appear in Cursor's MCP tools panel.

Connecting to Claude Desktop

This never worked. claude_desktop_config.json holds local servers only and ignores a URL without an error. Use the AgentPlaybooks plugin or a custom connector instead, with OAuth sign-in and no key — see AgentPlaybooks in Claude.

{
  "mcpServers": {
    "apb-my-assistant": {
      "transport": "http",
      "url": "https://apbks.com/api/mcp/YOUR_GUID"
    }
  }
}

Save this to your claude_desktop_config.json and restart Claude Desktop.

Connecting to Claude Code

One command:

claude mcp add apb-my-assistant https://apbks.com/api/mcp/YOUR_GUID --transport http

Verify with claude mcp list.

Authentication

Public playbooks require no authentication for read access. Anyone can connect and use the tools.

Private playbooks need an API key. Generate one from the Integrations tab and add it to your config:

{
  "mcpServers": {
    "my-playbook": {
      "url": "https://apbks.com/api/mcp/YOUR_GUID",
      "headers": {
        "Authorization": "Bearer apb_live_your_key_here"
      }
    }
  }
}

Write-back (saving to memory or canvas) always requires an API key, even for public playbooks. Keys come in three roles:

  • Viewer — Read-only
  • Coworker — Read + write
  • Admin — Full access

What Your AI Agent Gets

Once connected, your AI agent has access to:

| Component | MCP Capability | |---|---| | Skills | Accessible via list_skills / get_skill tools and Skills resource | | Memory | Read/write/search persistent key-value store | | Canvas | Read/write structured markdown documents | | Personas | System prompt and personality context | | Secrets | Server-side credential proxy (values never exposed) |

Built-in tools include read_memory, write_memory, search_memory, read_canvas, write_canvas, patch_canvas_section, get_canvas_toc, list_secrets, use_secret, and more.

Testing Your Connection

From the Integrations tab, copy the test command:

curl -s https://apbks.com/api/mcp/YOUR_GUID | head -c 200

If you see JSON with protocolVersion and serverInfo, you're good.

What's Next

  • Cursor Marketplace — We're working on listing AgentPlaybooks in the Cursor extension/MCP marketplace
  • Windsurf and other MCP-compatible IDEs — Same endpoint works everywhere
  • Management MCP Server — Shipped: see Update (October 2026) above for the account endpoint, with OAuth sign-in

Check the MCP Integration docs and Platform Integrations guide for the full reference.

AgentPlaybooks — Your AI's universal memory and toolkit. One playbook, every platform.

Share this article